Date:
Monday , January 07, 2002
Author:
Kyle Bennett
Google +1

Intel P4 Northwood CPU Review

The Pentium4 has changed once again, and this time the enthusiasts out there might just give a damn about it. Let's see how it stacks up against AMD's fastest 2000+ CPU.

continued...

The Test Systems:

Seeing how the AthlonXP 2000+ is launching today and there is no advancement in the technology, rather only in speed, we thought we would throw it into the mix for benchmarking. Also, instead of reviewing the 2.2GHz CPU, we decided on using the 2.0GHz. I figured that most of our brothers out there that would be purchasing one of these would most likely go with the cheaper of the two. If you can afford the 2.2GHz part, consider yourself lucky.

ABIT BD7-RAID: Intel Pentium 4 Willamette CPU clocked at a stock 2GHz (20 * 100FSB), Intel Pentium 4 Northwood CPU clocked at a stock 2GHz (20 * 100FSB) and also clocked at 2.52GHz (20 * 126FSB), 2 x 256MB Corsair XMS2400 DDR RAM, VisionTek Xtasy 6964 GF3 Ti500 VidCard, Netgear FA311 NIC, 30GB Western Digital ATA100 HD, Enermax 550 watt PSU.

VIA KT266A Reference Board: AMD AthlonXP 2000+ CPU clocked at 1.66GHz (12.5 * 133FSB), 2 x 256MB Corsair XMS2400 DDR RAM, VisionTek Xtasy 6964 GF3 Ti500 VidCard, Netgear FA311 NIC, 30GB Western Digital ATA100 HD, Enermax 550 watt PSU.

Overclocking:

Well, we've already let the cat out of the bag on our OCing success if you read the above specs. We picked the ABIT BD7-RAID, an i845 board, for our tests because some of its BIOS features cater to overclocking. In testing, it turned out just a bit slower than our Intel desktop board. Using the BD7-RAID gave us the opportunity to tweak the FSB, memory settings, Vcore, Vdimm, and also allowed us to "lock" the PCI bus to 33MHz so that it did not become a limiting factor. All Northwood testing shown is using the 2.0GHz Northwood CPU.

Article Image

While we did reach CPU clock speeds of over 2.6GHz, we found 2.52GHz to be extremely stable; we encountered no problems running the CPU at this speed at all. Amazingly enough, this level of OCing was not gained with any sort of super cooling, but rather with a very normal HSF that you will be finding for sale here in the States very soon.

Article Image   Article Image   Article Image

This is the first S478 CPU air cooler that we have seen with a copper core. Full testing has not been done yet to compare with our other S478 HSFs, but our preliminary data shows this CPU cooler to be leading the pack. It's made by a company called AVC in Taiwan and is codenamed Sun Flower, obviously because of its flower-like appearance.

Benchmarks:

SiSoft Sandra is a great all around performance benchmark when you need numbers immediately (no time to stand around waiting for hours on WebMark 2001 to finish). I find it best to run a few "preliminary" tests with SiSoft Sandra to give me a way to gauge overall performance before I dive into all out benchmarks.

We have not used this benchmark here recently, as we were not sure as to what we were seeing. We saw a tremendous increase in memory scores and thought it to be a fluke or bug. However, after talking at length with SiSoft and AMD, we are satisfied with the numbers the newer versions of Sandra are reporting. Special instructions and such are now being taken into consideration. Here is the full scoop from Mike Wall at AMD:

In reply to the request you made in the review, regarding the new Sandra memory benchmarks, I can give you some more info. Your comments about the new Sandra memory test are valid; the new scores are quite a lot higher than the old ones. Adrian may have more to say, but I can tell you that the new code is written to make the most efficient use of the hardware, and that's why it gets higher scores. It uses a "buffering" approach, and reads and writes data in blocks rather then frequently alternating read/write/read/write. And it uses the "streaming store" instructions for writing. This is the recommended way to get the best performance, at least on Athlon family. Software developers, including graphic card vendors, use these techniques to improve performance. You can read more details about it, in the Athlon Optimization Guide here

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/22007.pdf

Relevant parts of the guide are Chapter 5, p. 66 "Optimizing Main Memory Performance for Large Arrays" and also the sample code in Chapter 10, p. 180 which has Athlon-specific optimized memcpy() that works for any size memory block.

Also, the FAQ which installs as part of the Sandra application has a section about this:

Q: What's the deal with the new buffering SSE(2)/EMMX benchmarks?

A: In a nutshell, he new tests use the prefetching instructions to bring data into the CPU and store the data directly into memory bypassing the caches. In order maximise throughput, buffers are used to pre-fetch data into the caches so that it is already there when needed and to reduce switches between data streams.

Kyle, I hope that answers your questions about the new Sandra, at least in terms of the memory code and AMD platforms. Please feel free to send me any other questions on the subject.

-Mike

Article Image

Northwood 2.0GHz CPU

Article Image

Northwood 2.0GHz CPU

Article Image

Northwood 2.0GHz CPU

All of these numbers do tell a story. First of all, it tells us that the Intel Pentium4 still has piss poor FPU performance compared to the AthlonXP if SSE2 instructions are not utilized. However, when SSE2 is coded for in your software, you see that the FPU numbers are tremendous. When I ran the numbers for the DDR memory for the first time, I stared slack jawed at the monitor for a minute or so. I had truly thought that we would see Intel hobble the memory bandwidth on the i845 chipset to make their RAMBUS RDRAM look better. I think from these scores it is quite apparent that did not happen. With a 133MHz memory bus, anyone should be proud of these memory scores.

Now on to the OCed scores at 2.52GHz.

Article Image

Northwood 2.0GHz CPU at 2.52GHz (20 * 126MHz)

Article Image

Northwood 2.0GHz CPU at 2.52GHz (20 * 126MHz)

Article Image

Northwood 2.0GHz CPU at 2.52GHz (20 * 126MHz)

With our 25% overclock, it's no surprise to see a 25% increase in Whetstone and Dhrystone performance and the SSE2 instructions as well. But, yes, we are dealing with lower memory score, not by much, but still lower. This is because we had to clock down the memory to achieve the speeds we needed on the CPU bus and retain our tweaked memory settings. We also ran into some BIOS issues, but it's something we usually expect in sample boards.